Reward Learning from Multiple Feedback Types

Yannick Metz, András Geiszl, Raphaël Baur +1

Learning rewards from preference feedback has become an important tool in the alignment of agentic models. Preference-based feedback, often implemented as a binary comparison betwe…